AI as the New "Nuclear Club": Russian Tech Chief Urges Home‑Grown LLMs for National Security

By Elena Fabrichnaya and Gleb Bryanski

Alexander Vedyakhin, first deputy CEO of Sberbank, warned that artificial intelligence could confer influence comparable to nuclear technology for countries that seize an early lead. Speaking at Russia's AI Journey conference, he described a new global "nuclear club" defined by whether nations have their own national large language models (LLMs).

AI as a "nuclear project"

"AI is like a nuclear project," Vedyakhin said, arguing that sovereign states need domestic AI systems to protect sensitive data and maintain control over critical services. He urged Russia to develop at least two or three genuinely original models — not simply retrained foreign models — for use in online public services, healthcare and education.

"It is impossible to upload confidential information into a foreign model. It is simply prohibited. Doing so would lead to very unpleasant consequences."

Club membership and the global gap

Vedyakhin acknowledged that the United States and China currently lead by roughly six to nine months and said the front ranks are effectively closed to latecomers. He credited U.S. and Chinese firms with having abundant funding, specialists and computing power while noting Western sanctions make it harder for Russia to match leading-edge hardware and infrastructure.

He highlighted domestic efforts by Sberbank and other tech firms such as Yandex to catch up, saying Sberbank's GigaChat 2 MAX is comparable to ChatGPT 4.0 and its GigaChat Ultra Preview aligns with capabilities attributed to ChatGPT 5.0. The bank plans to open-source some newer models, including for commercial use.

Energy, cost and the risk of overinvestment

Developing advanced AI requires substantial computing and power. Vedyakhin estimated Russia's energy sector will need about 40 trillion roubles for generation and 5 trillion roubles for grids over the next 16 years to support growth in computing demand. He cautioned that high energy consumption can push returns on AI investments far into the future and warned against excessive infrastructure spending.

"We believe that excessive investments in AI infrastructure may indeed fail to pay off, given the rapid pace of technological development," he said, adding that Russia's more measured investment approach reduces the risk of an "AI bubble."

What could spark the next breakthrough?

Vedyakhin suggested two possible paths to the next major leap: a dramatic increase in LLM memory capacity or the emergence of an AI architecture that moves beyond today’s generative pre-trained transformer (GPT) paradigm. He pointed to innovations such as China's DeepSeek in 2024 as examples of how alternative approaches might shift the landscape.

While emphasizing skill and expertise as strengths — "What we can't achieve with sheer numbers, we achieve with skill" — Vedyakhin reiterated that closing the gap with global leaders will be difficult without access to comparable computing resources.